Controlled Acquisition and Abstention in Three-Channel Score Conflicts

Mengzhe Geng

cs.LGOct 7, 2026

Controlled Acquisition and Abstention in Three-Channel Score Conflicts

Authors: Mengzhe Geng

Organizations: National Research Council Canada

Abstract

When audio, video, and text disagree, accuracy alone does not show whether to acquire another source or abstain. We study these choices in a controlled three-score benchmark: a policy observes two signed scores, may request the third at a cost, and can abstain. The primary reward is mechanism-specific: abstention is correct only for one designated ambiguity mechanism and is penalized under mixed corruption. Matched controls show that a threshold policy matches always-request decisions with fewer requests; its advantage over always-answer fusion depends on the reward assigned to that ambiguity. On a partially held-out synthetic split, the threshold policy reaches 0.789 +/- 0.006 targeted decision accuracy and 0.481 +/- 0.014 utility across 83 seeds. A three-score majority reference reaches 0.626 +/- 0.008 and 0.252 +/- 0.016, but uses more information. In a matched-budget test, a train-only value selector improves utility over no-query and matched-random policies at 10% and 25% budgets, while pair uncertainty has higher utility at every budget. At 50% and 63.7% budgets, the selector lowers utility despite slightly higher non-ambiguous accuracy. If all abstentions are scored incorrect, majority outranks the threshold policy in utility. At a central temporal setting, full-trace controls match the neural models while position perturbations separate them. On held-out-actor emotion clips, eight-frame fusion has opposite-signed accuracy differences for two encoder pairs, with both actor intervals containing zero; matched-request routing gains are small and uncertain. These results separate full-modality accuracy from pre-request selection value and show that selection value depends on budget and the observed-pair ranking.

Figures & tables

Figure 1: The fixed decision rule. The policy requests text when the audio and video signs disagree or either observed score has magnitude below

0.35

. Without a request, the two strong, agreeing scores determine the answer. After acquisition, the policy answers only when at least two scores have magnitude at least

0.35

and share a sign; otherwise it abstains. Utility separately charges for a request and for non-ambiguous abstention.

Figure 2: Mean and standard deviation across 83 seeds. The partially held-out split compares full-information fusion references with policies that start from two scores and may request a third. Targeted decision accuracy uses the mechanism-conditioned target defined in the text. Utility includes a 0.10 query cost and a 0.20 cost for non-ambiguous abstention.

Method	Targeted acc. $\uparrow$	Non-ambig. binary acc. $\uparrow$	Utility $\uparrow$	Coverage	Query $\downarrow$	Abstain
Majority	$0.626$ $(0.008)$	$0.834$ $(0.007)$	$0.252$ $(0.016)$	$1.000$	$0.000$	$0.000$
Median	$0.626$ $(0.008)$	$0.834$ $(0.007)$	$0.252$ $(0.016)$	$1.000$	$0.000$	$0.000$
Weighted	$0.375$ $(0.007)$	$0.499$ $(0.008)$	$-0.251$ $(0.014)$	$1.000$	$0.000$	$0.000$
MLP	$0.415$ $(0.031)$	$0.553$ $(0.040)$	$-0.170$ $(0.061)$	$1.000$	$0.000$	$0.000$
Gated	$0.600$ $(0.063)$	$0.800$ $(0.083)$	$0.200$ $(0.125)$	$1.000$	$0.000$	$0.000$
Learned acquisition	$0.746$ $(0.010)$	$0.756$ $(0.009)$	$0.417$ $(0.019)$	$0.709$	$0.519$	$0.291$

Table 1: Partially held-out results over 83 seeds. Parentheses give sample standard deviations; rates are seed means with dispersion omitted. Mixed, burst, and ambiguity mechanisms are held out. Targeted accuracy rewards abstention only for designated ambiguity; non-ambiguous binary accuracy excludes ambiguity and counts abstentions as errors. Utility follows Equation 2 . Majority, median, weighted, MLP, and gated start with all three scores. Learned acquisition uses pair-only inputs, a full-score training target, and a different no-request action.

Method	Central targeted decision acc. $\uparrow$	Central util. $\uparrow$	Shared util. $\uparrow$	Hard util. $\uparrow$
Temporal fusion	$0.780$ $(0.011)$	$0.560$ $(0.021)$	$0.571$ $(0.013)$	$0.570$ $(0.016)$
Temporal gated	$0.786$ $(0.007)$	$0.571$ $(0.015)$	$0.567$ $(0.018)$	$0.572$ $(0.018)$
Full-trace ridge	$0.786$ $(0.006)$	$0.572$ $(0.012)$	$0.572$ $(0.012)$	$0.572$ $(0.014)$
Train-only template	$0.786$ $(0.009)$	$0.572$ $(0.018)$	$0.572$ $(0.018)$	$0.574$ $(0.013)$

Table 2: Equal-information temporal results at

T=64

, five seeds per setting. All methods receive full traces. Central and shared-permutation settings preserve the positional template; hard identity weakens signal and raises noise. Values are partially held-out targeted decision accuracy or utility, mean (sample standard deviation).

Appendix figures & tables13 assets

Supplementary material from the paper’s appendix.

Appendix

Method	Validation accuracy $\uparrow$	Test accuracy $\uparrow$
Majority / median	0.631	0.653
Weighted	0.587	0.616
MLP	$0.773$ $(0.002)$	$0.773$ $(0.002)$
Gated	$0.773$ $(0.002)$	$0.772$ $(0.002)$
Threshold	0.598	0.624

Appendix

Table 3: CMU-MOSEI descriptor accuracy on official folds. Five seeds; only learned models vary. These are not corruption-family splits.

Task head	Cost reduction $\uparrow$	95% interval
Modality-dropout ridge	$-0.0055$	$[-0.0126,\,0.0017]$
Separate-mask ridge	$-0.0041$	$[-0.0112,\,0.0032]$
Shared mask-aware ridge	$-0.0055$	$[-0.0128,\,0.0017]$

Appendix

Table 4: Projected CMU-MOSEI direct-action screen at a 0.50 request budget. Mean normalized cost reduction versus no request over five seeds; paired video-group intervals use 10,000 bootstrap draws. Positive values favor acquisition.

Input	Target (%) [CI] $\uparrow$	Donor S/O (%) $\uparrow$	Pair S/O (%) $\uparrow$	Gain T/S/O (%) $\uparrow$
Audio	30 [15, 52]	50 / 55	25 / 15	0 / 10 / 5
Video	40 [22, 61]	40 / 55	20 / 20	5 / 0 / 5
A+V	40 [22, 61]	40 / 55	20 / 20	5 / 0 / 5

Appendix

Table 5: Perception Test source-feasibility results (percent; target shows Wilson 95% CI; donor/pair rates use 20 pairs). A+V is aligned input; T/S/O is target/same/opposite. Minima: 70 source accuracy, 50 pair support, and an absolute accuracy gain of 20%.

Visual frames	Video acc. (%) $\uparrow$	A+V acc. (%) $\uparrow$	A+V $-$ audio absolute diff. (%) $\uparrow$
1	22.03	35.45	$-9.65$ [ $-13.70$ , $-5.28$ ]
4	37.78	48.15	$+3.05$ [ $-2.80$ , $+8.21$ ]
8	43.09	50.96	$+5.87$ [ $-0.56$ , $+11.77$ ]

Appendix

Table 6: CREMA-D actor-held-out singleton evaluation with frozen wav2vec 2.0 and Vision Transformer (ViT) features. A+V denotes audio–video fusion. Every condition uses the same 45/15/15/16 actor roles and 1,244 evaluation clips; only the number of uniformly sampled visual frames changes. The last column is the absolute accuracy difference (A+V minus audio) computed from unrounded scores; intervals use 10,000 paired bootstrap resamples of the 16 evaluation actors.

Frozen pair	Audio (%) $\uparrow$	A+V (%) $\uparrow$	A+V $-$ audio (%) $\uparrow$	Request (%) $\downarrow$	Router $-$ matched random absolute diff. (%) $\uparrow$
wav2vec 2.0–ViT	45.10	50.96	$+5.87$ [ $-0.56$ , $+11.77$ ]	99.36	$+0.04$ [ $-0.30$ , $+0.36$ ]
HuBERT–DeiT	58.92	56.43	$-2.49$ [ $-8.40$ , $+2.81$ ]	71.70	$+0.42$ [ $-1.02$ , $+1.87$ ]

Appendix

Table 7: Pinned CREMA-D comparison on 1,244 singleton-vote clips from 16 held-out actors. Pair names list audio then video encoders; A+V uses both. Audio and A+V columns report accuracy, and request is the realized audio-to-video router rate. The two difference columns report A+V minus audio and router minus matched-random accuracy, respectively, computed from unrounded scores as absolute differences (%). Brackets give 95% actor bootstrap intervals; matched random uses the same request rate.

Frozen pair	Full headroom (%) $\uparrow$	10% $\bm{\Delta}$ (%) $\uparrow$	25% $\bm{\Delta}$ (%) $\uparrow$	50% $\bm{\Delta}$ (%) $\uparrow$	75% $\bm{\Delta}$ (%) $\uparrow$
wav2vec 2.0 + ViT	$+6.28$	$+0.54$	$+1.24$	$+1.04$	$+0.89$
wav2vec 2.0 + DeiT	$+3.09$	$+0.01$	$+0.16$	$+0.34$	$+0.05$
HuBERT + ViT	$+0.89$	$+0.14$	$-0.13$	$+0.25$	$+0.68$
HuBERT + DeiT	$-2.35$	$+0.06$	$+0.34$	$+0.71$	$+0.47$

Appendix

Table 8: CREMA-D comprehensive frozen-feature grid. Values are means over six overlapping pre-label actor reallocations. Column percentages are calibration targets, not exact evaluation request rates. Full headroom is full audio–video accuracy minus audio-only accuracy.

\Delta

is policy accuracy minus expected matched-random accuracy at the realized request rate; all values are absolute accuracy differences (%). The reallocation means are descriptive split robustness, not independent-population estimates.

Frozen pair	Full headroom	10%	25%	50%	75%
wav2vec 2.0 + ViT	6/6	6/6	6/6	5/6	5/6
wav2vec 2.0 + DeiT	5/6	3/6	3/6	5/6	5/6
HuBERT + ViT	5/6	3/6	3/6	3/6	3/6
HuBERT + DeiT	1/6	3/6	4/6	5/6	5/6

Appendix

Table 9: Sign consistency of the CREMA-D comprehensive grid. Each cell gives the number of positive reallocation-level estimates out of six overlapping actor allocations. Full headroom is full audio–video accuracy minus audio-only accuracy; the four request columns give policy accuracy minus matched-random accuracy at the realized request rate. Counts summarize split sensitivity and are not confidence intervals or independent-population estimates.

Team	Records	Class counts	Accuracy (%) $\uparrow$
			A	V	T	A+V	A+T	V+T	A+V+T
ES2006	123	9 / 23 / 11 / 80	65.04%	27.64%	82.11%	52.85%	64.23%	34.96%	21.95%
ES2011	121	9 / 23 / 10 / 79	56.20%	19.83%	70.25%	36.36%	53.72%	37.19%	49.59%

Appendix

Table 10: AMI accuracy by evaluation meeting team. Class counts follow the observed target-class order elicit , minor , other , task . Accuracy is the percentage correct within each team for the indicated feature subset; A, V, and T denote audio, video, and text. The two-team macro is the unweighted mean of the ES2006 and ES2011 values; the breakdown is descriptive because only two evaluation teams are available.

Team	Policy	Accuracy (%) $\uparrow$
ES2006	Audio–video static	52.85%
	Learned text router	48.78%
	Matched random	44.72%
	Matched confidence	48.78%
	Label-informed oracle	69.92%
ES2011	Audio–video static	36.36%

Appendix

Table 11: AMI router accuracy by evaluation team for the available audio–video baseline and text-request policies. Each percentage is computed within the team’s records. Each exact-count control uses 38 requests across all 244 records; team allocations can differ. Macro differences described in the text are unweighted means of the two team-level differences.

Policy	Targeted acc. $\uparrow$	Utility $\uparrow$	Query $\downarrow$	Coverage
Pair-selective	$0.594$ $(0.006)$	$0.110$ $(0.012)$	$0.000$	$0.366$
Query-majority	$0.628$ $(0.005)$	$0.192$ $(0.010)$	$0.634$	$1.000$
Always-query-selective	$0.790$ $(0.003)$	$0.447$ $(0.007)$	$1.000$	$0.594$
Threshold	$0.790$ $(0.003)$	$0.484$ $(0.006)$	$0.634$	$0.594$

Appendix

Table 12: Matched synthetic controls at

\sigma=0.22

\tau=0.35

. Targeted decision accuracy and utility are mean (sample standard deviation) over five paired seeds. All methods start with the same pair; all requests cost 0.10.

Query budget	Train-only value $\uparrow$	Pair uncertainty $\uparrow$	Matched random $\uparrow$	Oracle upper bound $\uparrow$	Value–random $\Delta$ U (95% CI) $\uparrow$
10%	$-0.0321$ $(0.0166)$	$-0.0304$ $(0.0164)$	$-0.0570$ $(0.0168)$	$0.1479$ $(0.0159)$	$+0.0249$ $[0.0233,\,0.0264]$
25%	$-0.0418$ $(0.0161)$	$-0.0296$ $(0.0166)$	$-0.0721$ $(0.0164)$	$0.2455$ $(0.0151)$	$+0.0303$ $[0.0280,\,0.0325]$
50%	$-0.1254$ $(0.0175)$	$-0.0081$ $(0.0147)$	$-0.0994$ $(0.0164)$	$0.2205$ $(0.0151)$	$-0.0260$ $[-0.0292,\,-0.0228]$
63.7%	$-0.1952$ $(0.0167)$	$-0.0946$ $(0.0151)$	$-0.1148$ $(0.0158)$	$0.2068$ $(0.0151)$	$-0.0804$ $[-0.0832,\,-0.0776]$

Appendix

Table 13: Matched-budget selector comparison under one shared terminal rule over 83 seeds. Cells show utility mean (sample standard deviation). The final column gives the train-only value selector’s paired mean utility difference from matched random with a 95% paired-seed percentile interval. The no-query reference is

-0.0459

(

0.0160

) at each budget.

Figure 3: Cost sensitivity on the partially held-out split. Curves show the active policy’s mean utility as query cost varies; shaded regions are one standard deviation across seeds. The dashed line is the majority baseline, whose query and abstention rates are zero. The star marks the campaign costs.

Costs $(\lambda_{q},\lambda_{a})$	Threshold utility $\uparrow$	Majority utility $\uparrow$	Utility difference $\uparrow$
$(0.00,0.00)$	$0.578$ $(0.012)$	$0.252$ $(0.016)$	$0.326$
$(0.10,0.20)$	$0.481$ $(0.014)$	$0.252$ $(0.016)$	$0.230$
$(0.20,0.40)$	$0.385$ $(0.015)$	$0.252$ $(0.016)$	$0.133$
$(0.40,0.80)$	$0.192$ $(0.018)$	$0.252$ $(0.016)$	$-0.059$

Appendix

Table 14: Cost sensitivity in the 83-seed scalar study. Policy utilities are mean (sample standard deviation); the utility-difference column is threshold utility minus majority utility. Majority is unchanged because it never requests or abstains. Costs are query

\lambda_{q}

and non-ambiguous abstention

\lambda_{a}

Explore similar work

CHASE: Competing Hypotheses for Ambiguity-Aware Selective Prediction
May 2, 2026Kartik Jhawar, Yuhao Geng, Atul N. Parikh +1Selective Prediction Selective Inference
Institute for Digital Molecular Analytics and Science, NTU, Singapore 636921 · School of Electrical and Electronic Engineering, NTU, Singapore 639798 · School of Materials Science and Engineering, NTU, Singapore 639798 +1
Benchmarking candidate coverage and rejection policy transfer in typed decision models
Oct 2, 2026Jiawen Lu, Tongtong WuLLM Evaluation Selective Prediction
Monash University
Selective QA over Conflicting Multi-Source Personal Memory: A Diagnostic Testbed and Method Comparison
May 28, 2026Tiancheng Yang, Matthias Schonlau, Ilia SucholutskyConversational Memory Question Answering

May 2, 2026, cs.CVJ/K move · Enter open · S save

CHASE: Competing Hypotheses for Ambiguity-Aware Selective Prediction

Kartik Jhawar, Yuhao Geng, Atul N. Parikh, Lipo Wang

Institute for Digital Molecular Analytics and Science, NTU, Singapore 636921 · School of Electrical and Electronic Engineering, NTU, Singapore 639798 · School of Materials Science and Engineering, NTU, Singapore 639798+1

Selective Prediction Selective Inference Decision-Making under Uncertainty

Open paper AskPDF arXiv

Standard selective prediction methods typically estimate uncertainty from the output of a single predictive branch. While effective for general uncertainty estimation, these approaches often struggle under partial observability, where local temporal evidence can be contradictory and standard confidence scores become misleading. We introduce CHASE (Competing Hypotheses for Ambiguity-Aware Selective Prediction), a selective prediction framework that explicitly compares structured temporal explanations to determine whether to commit to a decision or abstain. Because genuine ambiguity causes the score gap between competing hypotheses to collapse, CHASE optimizes a ranking-aware selector over these hypothesis margins to globally separate safe commitments from fundamentally uncertain ones. We evaluate this framework on the problem of hidden connectivity inference, utilizing a controlled, physically grounded simulator inspired by the dynamics of giant unilamellar vesicles (GUVs), alongside zero-shot qualitative transfer (without retraining or fine tuning) to representative real GUV videos. Our experiments demonstrate that explicitly reasoning over competing hypotheses provides a superior balance of metrics. Compared to canonical uncertainty baselines, CHASE achieves statistically significant gains in overall no-abstain accuracy, three-way accuracy, and overall ambiguity-aligned abstention (at 80% coverage). Specifically, it yields up to an 11.0% relative mean improvement in overall alignment, alongside up to an 8.8% relative boost in three-way accuracy in the very-high ambiguity regime. By maintaining a selective risk boundary strictly at par with the best baselines at 80% coverage, and reducing overall risk by 9.9% at 90% coverage, this framework offers a more reliable approach to decision-making under structured ambiguity.

Abstract

Explore similar work

CHASE: Competing Hypotheses for Ambiguity-Aware Selective Prediction

Benchmarking candidate coverage and rejection policy transfer in typed decision models

Selective QA over Conflicting Multi-Source Personal Memory: A Diagnostic Testbed and Method Comparison